| name | create-command |
| description | Create slash commands for Claude Code. USE WHEN user says 'create a command', 'make a command', 'add a command', 'new slash command', 'build a command for', 'I need a shortcut for', 'make a quick action', 'automate this action'. |
Create Command
Teaches Claude how to create properly structured slash commands through a guided interview process.
What is a Command?
A command is a shortcut you type to trigger an action. Commands live in .claude/commands/ and are invoked with /project:[command-name].
Key insight: Commands are for quick, repeatable actions. Skills are for complex workflows.
Example: /project:status runs the status command.
The 3-Phase Interview Process
When creating a NEW command, guide the user through these phases:
Phase 1: ACTION
Ask the user:
- "What action should this command trigger?"
- "What's the end result?"
- "Is this something you'll run often?"
Goal: Understand the repeatable action
Phase 2: ARGUMENTS
Ask the user:
- "Does this command need input from you each time?"
- "What information would you pass to it?"
Goal: Determine if $ARGUMENTS or $1, $2 are needed
Phase 3: TOOLS
Ask the user:
- "What does this command need to do?" (read files, run git, search web, etc.)
- "Should it auto-approve these tools or ask permission?"
Goal: Determine allowed-tools for smooth execution
Command Structure
Commands are single markdown files:
.claude/commands/
├── status.md → /project:status
├── research.md → /project:research
└── deploy.md → /project:deploy
Namespaced commands:
.claude/commands/
└── git/
├── status.md → /project:git:status
└── sync.md → /project:git:sync
Command Template
---
description: What this command does (shown in command menu)
argument-hint: [optional hint for arguments]
allowed-tools: Tool1, Tool2, Tool3
---
Your command instructions here.
Use $ARGUMENTS to access what the user typed after the command.
Use $1, $2, $3 for individual positional arguments.
Use @filename to include file contents.
YAML Frontmatter Options
| Field | Required | Purpose |
|---|
description | Yes | Shown when user types /project: - keep it brief |
argument-hint | No | Hint for arguments, e.g., [video URL] or <topic> |
allowed-tools | No | Auto-approve these tools (no permission prompts) |
model | No | Override model: haiku, sonnet, opus |
Variables
| Variable | What it contains | Example |
|---|
$ARGUMENTS | Everything typed after the command | /project:research AI tools → AI tools |
$1, $2, $3 | Individual positional arguments | /project:compare file1.md file2.md → $1=file1.md, $2=file2.md |
@filename | Include file contents inline | @CLAUDE.md embeds the file |
!`command` | Embed bash output (context section) | !`git status` shows current git status |
Allowed Tools
Auto-approve specific tools so Claude doesn't ask permission:
allowed-tools: Read, Write, Bash(git:*)
Common patterns:
| Pattern | What it allows |
|---|
Read | Read any file |
Write | Write/create files |
Edit | Edit existing files |
Glob | Search for files by pattern |
Grep | Search file contents |
Bash(git:*) | Git commands only |
Bash(npm:*) | NPM commands only |
Bash(python:*) | Python scripts only |
WebSearch | Web searches |
WebFetch | Fetch web pages |
Skill | Invoke other skills |
Task | Spawn sub-agents |
Security tip: Only auto-approve what's needed. More restrictive = safer.
Examples
Simple Command (No Arguments)
---
description: Check git status and recent commits
allowed-tools: Bash(git:*), Read
---
Show me:
1. Current git status
2. Last 5 commits
3. Any uncommitted changes
Usage: /project:status
Command with Arguments
---
description: Research a topic thoroughly
argument-hint: [topic]
allowed-tools: WebSearch, WebFetch, Read
---
Research this topic: $ARGUMENTS
Provide:
- Key concepts explained simply
- Recent developments (2024-2025)
- Practical applications for business
- 3 actionable insights
Usage: /project:research AI automation for small business
Command with Positional Arguments
---
description: Compare two files
argument-hint: [file1] [file2]
allowed-tools: Read
---
Compare these two files:
- File 1: $1
- File 2: $2
Show:
1. Key differences
2. What changed
3. Which is better and why
Usage: /project:compare old-version.md new-version.md
Command with File Reference
---
description: Review code changes against project standards
allowed-tools: Bash(git:*), Read
---
Review the current code changes.
Project standards to follow:
@CLAUDE.md
Focus on:
1. Does it follow our patterns?
2. Any bugs or issues?
3. Suggestions for improvement
Usage: /project:review
Command with Embedded Bash Context
---
description: Smart commit with context
allowed-tools: Bash(git:*)
---
## Current State
- Status: !`git status --short`
- Branch: !`git branch --show-current`
- Recent: !`git log --oneline -3`
Based on the changes above, create a clear commit message and commit.
Usage: /project:commit
Command that Delegates to Skill
---
description: Transform article into social content
argument-hint: [article URL or path]
allowed-tools: Skill, Read, WebFetch
---
Use the `write` skill to transform this article into:
1. Twitter thread
2. LinkedIn post
3. 3 quote posts
Article: $ARGUMENTS
Usage: /project:transform https://example.com/article
Creation Checklist
Before finalizing a command, verify:
When to Use Commands vs Skills
| Use a Command when... | Use a Skill when... |
|---|
| Quick, repeatable action | Complex multi-step workflow |
| Same action, different inputs | Needs interview/discovery |
| "Just do this thing" | "Help me figure out how to..." |
| Single focused task | Multiple possible paths |
Examples:
/project:status → Command (quick action)- "Create a youtube research skill" → Skill (complex, needs setup)
Common Mistakes
- No description - Always include one (it's shown in the menu)
- Forgetting $ARGUMENTS - Use it if you expect user input
- Over-permissive tools - Only auto-approve what's needed
- Too complex - If it needs multiple phases, make it a skill instead
- Vague instructions - Be specific about what Claude should do
Location: .claude/commands/[command-name].md
Invoke: /project:[command-name]
Namespaced: /project:[folder]:[command-name]
